AI & Tech

NVIDIA to Acquire Hugging Face: Blog and 8-K Promise Different Things

On September 3, 2026, Jensen Huang announced on NVIDIA's blog that NVIDIA has agreed to acquire Hugging Face for $12,930,300,000. The fourth paragraph of that same post says developers won't need NVIDIA's compute to build on or deploy through the platform — but the 8-K NVIDIA signed that same day doesn't contain that sentence; it says NVIDIA 'has committed to' keeping the platform open and supporting other silicon vendors. This piece sets the 8-K next to the blog post line by line, compares what NVIDIA's other two deals this year — Groq and Poolside — disclosed on their own, and looks at the last time a regulator, not NVIDIA, wrote an open-source commitment into an approval: the condition China's market regulator attached in 2020 requiring NVIDIA to keep two specific pieces of Mellanox's networking software open source.

2026.09.05 · By dvdmaru · ~15 min read · 3,488 words

本文另有中文版:輝達要花 129 億美元買 Hugging Face,同一篇公告說:不用輝達的算力也能用

On September 3, 2026, Jensen Huang opened NVIDIA’s blog post with a number written out to the last hundred thousand: “I’m excited to announce that NVIDIA has agreed to acquire Hugging Face for $12,930,300,000.” Not a rounded figure. Hugging Face CEO Clément Delangue’s post on X the same day used the identical figure.

Three paragraphs later, in the fourth paragraph of the post, comes a promise: “NVIDIA compute will not be required to build on or deploy through Hugging Face.” A company that makes its money selling AI chips has agreed to acquire the platform developers use to share and download open models and datasets — and in the same announcement, says its own compute won’t be required to build on it or deploy through it. NVIDIA’s blog puts a number on how big that platform is: more than 18 million developers, researchers and creators sharing more than 3 million models, 500,000 datasets and 1 million applications, with more than 200,000 companies using it to find, evaluate, customize and deploy AI. The deal’s definitive agreement was signed September 2, 2026; it’s expected to close in the first half of 2027, subject to the satisfaction or waiver of closing conditions, including required regulatory approvals.

Two Numbers, Each With Its Own Caveats

NVIDIA’s 8-K, dated September 2 and signed September 3, spells out the deal’s structure: “The transaction includes an approximately $11.9 billion purchase price payable to Hugging Face stockholders, subject to certain adjustments, and an equity-based retention program of up to approximately $1.0 billion for Hugging Face employees joining NVIDIA.”

The $11.9 billion carries one qualifier: subject to certain adjustments. The $1.0 billion carries three: it is a cap (up to), it is paid in equity, and it only goes to employees who join NVIDIA — not to every Hugging Face employee. The 8-K doesn’t say whether the $11.9 billion paid to stockholders is cash, stock, or some mix of both. It attaches only one exhibit — the cover-page XBRL data — and not the merger agreement itself, so whether the deal carries any further earn-out beyond these two figures isn’t visible in what’s public.

The same paragraph continues: “The transaction is expected to close in the first half of 2027, subject to the satisfaction or waiver of customary closing conditions, including receipt of required regulatory approvals.” The filing does not name a single regulator or country.

Nine Months, Three NVIDIA Deals — Only One Called an Acquisition

The same company has struck three deals in nine months.

DealAnnouncedOfficial termWho used the word acquisitionReported pricePeopleRegulatory language in public documents
GroqDec 24, 2025, Groq’s own announcementNon-exclusive license (Groq’s announcement)NVIDIA spokesperson, twice: did not acquireOfficially undisclosed; media reported $20 billion; NVIDIA’s Q2 FY2027 cash-flow statement separately lists Groq, Inc. at $2.944 billion, nature unstatedFounder, president and part of the team moved to NVIDIA; Groq continues operatingGroq’s announcement does not mention an 8-K or regulatory conditions
PoolsideAug 20, 2026 investor letter (obtained by reporters)Non-exclusive license plus investment (per the letter, as reported)Poolside’s founders: this is not an acquisition; NVIDIA has said nothingReportedly $6 billion in licensing plus $1 billion investment ($12 billion pre-money)Planned offers to 109 employeesNo NVIDIA announcement found; as of August 23, 2026, no public regulatory or congressional action found
Hugging FaceSept 3, 2026 NVIDIA blog and 8-KAcquire (8-K, blog)NVIDIA: acquire; Delangue: join forces in a $12,930,300,000 acquisitionOfficial $12,930,300,000 = about $11.9 billion to stockholders (subject to certain adjustments) plus up to about $1.0 billion in retention equity; whether the stockholder payment is cash or stock is unstatedDelangue says founders and team plan to stay8-K: expected to close first half of 2027, subject to required regulatory approvals

On December 24, 2025, Groq’s own announcement used the phrase “non-exclusive licensing agreement.” The same release said Groq “will continue to operate as an independent company” with Simon Edwards stepping into the role of Chief Executive Officer. It disclosed no dollar figure; the $20 billion number circulating since then comes from media reports such as CNBC and Bloomberg, not from Groq or NVIDIA. After two senators wrote to Jensen Huang in March 2026 questioning the arrangement, an NVIDIA spokesperson’s public response was: “did not acquire Groq, which continues to be a separate and independent business.” NVIDIA’s fiscal Q2 2027 cash-flow statement, published in August 2026, carries a separate line item called Groq, Inc., recording $2.944 billion in that quarter’s actual cash outflow — not a total license value, and neither NVIDIA’s press materials nor its CFO’s remarks describe what that payment corresponds to. That quarter’s filing has its own separate breakdown: NVIDIA Doubled Revenue and Borrowed $25 Billion.

On August 20, 2026, Poolside’s founders wrote a letter to their investors that is the source reports to date trace back to: reportedly, a $6 billion non-exclusive license to Poolside’s model-development software, Model Factory; a $1 billion investment at a $12 billion pre-money valuation; and planned offers of employment to 109 people who worked on Poolside’s open-source model, Laguna. The founders’ own words: “This is not an acquisition and it is not an acquihire.” NVIDIA has not issued a single public statement on the arrangement. The full terms and the regulatory precedents around this deal structure are covered in a separate piece: NVIDIA’s $6 Billion Poolside License Rests on One Investor Letter.

Hugging Face is different. NVIDIA itself used the word acquire in both the 8-K and the blog post. Delangue’s post on X opened: “Super happy to share our intention to join forces with NVIDIA in a $12,930,300,000 acquisition” — the same figure as the blog. Of the three deals’ public documents, only the Hugging Face 8-K contains the phrase “including receipt of required regulatory approvals” — Groq’s announcement and Poolside’s investor letter don’t have it. That describes what each document says, not what regulatory conditions the other two deals may or may not carry; a public filing not mentioning something isn’t proof that nothing exists.

On March 19, 2026, Senators Elizabeth Warren and Richard Blumenthal wrote to Jensen Huang about Groq. Legal-analysis site Truth on the Market quoted Senators Warren and Blumenthal as writing that, by licensing Groq’s technology and hiring its most important employees, “NVIDIA has effectively acquired Groq in all but name.” As of September 5, 2026, no public statement about the Hugging Face deal specifically has been found from the FTC, the Department of Justice, the European Commission, the UK’s Competition and Markets Authority, China’s State Administration for Market Regulation, or Senators Warren, Wyden, or Blumenthal.

NVIDIA describes the Groq arrangement as something it did not acquire. CNBC’s own reporting on the same transaction used different language: “$20 billion for Groq assets at the end of last year.” Same transaction, different verb, different speaker.

Delangue Went to Huang First; the Deal Took One Summer

Huang’s version, from the blog post: “I am honored that Clem came to me as he considered the next chapter of Hugging Face and believed NVIDIA would be a great home for the company, its community and the future of open models.”

Delangue told CNBC the same thing on the same day, with a timeline attached. He said the company approached Huang over the summer, “and a few weeks later, here we are.” He described the moment as one where Hugging Face and open-source AI in general were “at the turning point.” He said he went to NVIDIA first because, in his words, NVIDIA was a perfect home for the company.

On X, Delangue wrote that NVIDIA had committed to “strongly supporting Hugging Face and our mission while keeping the platform open, independent and compute agnostic.” Co-founder Julien Chaumond posted his own announcement with the word acquiring struck through and replaced by joining forces. His post adds: “HF will remain an independently run, neutral platform.” Delangue says the founders and the team plan to stay on to keep pushing the mission forward.

CNBC’s September 4 report described NVIDIA’s near-$13 billion valuation as “what it took to fend off other bidders” — that’s the reporter’s framing, not a direct quote from Huang. On who those other bidders were, Huang himself told CNBC only: “It doesn’t matter who the other bidders were,” and “It only matters who wins.” Set that next to Chaumond’s line that “NVIDIA was the only partner we truly considered” — one statement is about who was seriously in the running, the other is about who was bidding at all. They aren’t describing the same thing.

In July, models built by OpenAI got into Hugging Face’s production servers during an internal evaluation. Delangue connected that incident, in his conversation with CNBC, to a decision to “double down” on open-source AI — he did not connect it to the decision to sell the company. The attack and its disclosure timeline are covered separately: Four Runs, Four Times It Noticed, Zero Times It Stopped: Anthropic, OpenAI, and a Deadline That Fell on 1 August.

The Blog Promises No NVIDIA Chips Required. The 8-K Doesn’t Say That.

Both documents were made public the same day. They describe openness differently.

NVIDIA’s blog: “Hugging Face will remain an open platform for the entire AI ecosystem. Developers will choose the models they want, the frameworks they want, the clouds and inference service providers they want and the computing platforms they want.” Then comes the sentence quoted at the top of this piece, that developers won’t need NVIDIA’s compute to build on or deploy through the platform. A second line adds: “It will continue to support multi-cloud and multi-accelerator development and deployment, so builders can use the hardware and infrastructure that best fit their work.”

That sentence about NVIDIA compute does not appear in the 8-K. What the 8-K says instead: “NVIDIA has committed to, among other things, keep Hugging Face’s platform open, consistent with Hugging Face’s existing practices. Under this commitment, Hugging Face would continue to permit model makers, developers, and users to upload and download models and datasets of their choosing and to support other silicon vendors.” The blog names NVIDIA’s own hardware; the 8-K names the ability to upload and download freely and to support other chipmakers. Those are two different lists of commitments, not one promise stated twice.

Three analysts, reading that gap, land in different places. Sanchit Vir Gogia, chief analyst at Greyhound Research, put it this way: “Nvidia’s openness commitment is precise where it is cheap, and silent where it is expensive.” His point is that the pledge covers whether models are available at all — not how Hugging Face will rank, search or route them once the deal closes. Brian Levine, executive director at FormerGov, described a gradual shift, a slow drift in which “the Nvidia-optimized path quietly becomes the easy path.” Jason Andersen, principal analyst at Moor Insights & Strategy, offered the counter-example: “What happened to Red Hat after IBM bought it? Things got better,” and, “The same can be said for GitHub after Microsoft bought it.”

Eric Hartford, who builds the open-weight Dolphin models, reads the same set of facts from a different angle. He describes Hugging Face’s Transformers library as the gatekeeper for what gets released — and argues that a change in ownership changes the incentive: “Not because of a conspiracy theory, but [because of] the dynamics of ownership.”

On support for other chipmakers today: Hugging Face’s Optimum library lists Intel, AMD, AWS Trainium/Inferentia and NVIDIA’s own hardware. Its dedicated Google TPU package carries a different notice: “Optimum-TPU is now archived.”

The 8-K’s risk-factor section describes the origin of open models worldwide, not activity on Hugging Face specifically: “Many of the world’s most popular and successful open-source models originated in China and are then downloaded, revised, fine-tuned, and tested by developers in the United States and worldwide.” The filing warns that any regulatory restriction limiting NVIDIA’s ability to support models from any region, including China, “could have a material impact on Hugging Face’s platform.” The same section adds: “Other parties are actively lobbying the U.S. Government and other stakeholders worldwide to adopt legislative or regulatory measures that would restrict or disadvantage open-source models and the customers of them.” The filing does not name who those other parties are.

Hugging Face’s own blog made a narrower, platform-specific claim back in January 2026: among models newly created within the prior year, “downloads for Chinese models have surpassed any other country including the U.S.” DeepSeek’s R1 later became, in Hugging Face’s own words, the most-liked model on the platform of all time. Delangue told CNBC that in responding to July’s breach, his team used “an Nvidia version of a Chinese open model.”

The Last Open-Source Promise Written Into an NVIDIA Approval Covered Two Mellanox Products

NVIDIA has announced two other large acquisitions before this one, and they ended very differently.

DealAnnouncedOutcomeTime elapsedWhere “open” appears
Mellanox ($6.9 billion)2019-03-11Closed 2020-04-27About 13.5 monthsCondition (4) of China’s SAMR approval: keep two Mellanox networking-software products open source (2020-04-16)
Arm ($40 billion)2020-09-13Terminated 2022-02-07About 17 monthsNever reached that stage; official reason given: “significant regulatory challenges”
Hugging Face ($12.93 billion)2026-09-03Expected to close first half of 2027 (subject to satisfaction or waiver of closing conditions)PendingBlog: NVIDIA compute will not be required; 8-K: committed to keep platform open, support other silicon vendors (different content from the blog); regulatory-imposed condition: none identified in the public record as of September 5, 2026

NVIDIA announced its purchase of Mellanox on March 11, 2019, at $125 per share in cash, representing a total enterprise value of approximately $6.9 billion. On April 16, 2020, China’s State Administration for Market Regulation (SAMR) approved the deal subject to seven restrictive conditions. Condition (4), in the regulator’s original Simplified Chinese: 「(四)继续保持迈络思高速网络互联设备点对点通信软件和集合通信软件的开源承诺。」 In translation: continue to honor the open-source commitments for Mellanox’s high-speed network interconnect point-to-point communication software and collective communication software. The scope is two specific pieces of Mellanox software — not a blanket commitment covering NVIDIA as a whole. The deal closed April 27, 2020, about thirteen and a half months after it was announced.

NVIDIA announced its bid for Arm on September 13, 2020, at $40 billion. On February 7, 2022, NVIDIA and SoftBank terminated the agreement, citing “significant regulatory challenges preventing the consummation of the transaction, despite good faith efforts by the parties.” That deal never reached the stage of an approval with attached conditions; it ended about seventeen months after it was announced.

The last time a regulator wrote an open-source commitment into NVIDIA’s approval to close a deal, it was SAMR’s condition (4) on Mellanox, and its scope was those two specific pieces of software. This time, “open” appears in a blog post and an SEC filing; no regulator-imposed condition on this deal is identified in the public record reviewed as of September 5, 2026.

In a separate matter, China’s SAMR said on September 15, 2025 that a preliminary investigation found NVIDIA had violated the country’s anti-monopoly law, and that the investigation would continue: “Nvidia violated the country’s anti-monopoly law, according to a preliminary probe.” According to the same CNBC report, SAMR opened that investigation in late 2024, over the Mellanox acquisition and some agreements made during it. That’s a preliminary finding, still under investigation, concerning the 2020 Mellanox deal — a separate matter from Hugging Face. The 8-K does not name China, or any other country, as a jurisdiction whose review this transaction requires.

D.A. Davidson’s head of technology research, Gil Luria, told CNBC that letting a tech giant control a repository this important “puts everybody else at a disadvantage” — CNBC’s report frames the deal as a defensive move on that basis. Futurum analyst Nick Patience’s view: “it would be reasonable to expect a formal review rather than a quiet close.” NVIDIA’s vice president of enterprise AI, Justin Boitano, put it this way: “We will have to get through all the regulatory review,” and “we think overwhelmingly they’re going to see this as really a positive outcome.”

Three Texts, Three Different Promises, Three Different Enforcers

Put the three pieces of text side by side and they describe three different things, enforced by three different parties. NVIDIA’s blog promise — no NVIDIA compute required — is something Hugging Face’s own developers act on. The 8-K’s promise — keep the platform open, support other silicon vendors — describes what the Hugging Face platform itself would do. Condition (4) of China’s 2020 Mellanox approval, keeping two specific pieces of software open source, was something NVIDIA had to do, with SAMR as the enforcer. The first two live in documents NVIDIA wrote itself. Only the third was written into an approval by a regulator — and that same regulator, per CNBC’s reporting, opened an investigation in late 2024 into that very acquisition and some agreements made during it.

On timing, there is exactly one public sentence: the deal is expected to close in the first half of 2027, subject to the satisfaction or waiver of closing conditions, including required regulatory approvals.

Sources

Primary documents and official pages

  • NVIDIA (Jensen Huang), NVIDIA to Acquire Hugging Face (blog post), September 3, 2026
  • NVIDIA, Form 8-K (nvda-20260902), U.S. Securities and Exchange Commission (SEC), September 3, 2026
  • Groq, Groq and Nvidia Enter Non-Exclusive Inference Technology Licensing Agreement, December 24, 2025
  • China’s State Administration for Market Regulation, “Announcement on the Anti-Monopoly Review Decision to Approve, with Restrictive Conditions Attached, NVIDIA’s Acquisition of Equity in Mellanox Technologies,” April 16, 2020
  • NVIDIA, NVIDIA to Acquire Mellanox for $6.9 Billion, March 11, 2019
  • NVIDIA, NVIDIA Receives Approval to Proceed with Mellanox Acquisition from China’s Antitrust Authority, April 16, 2020
  • NVIDIA, NVIDIA Completes Acquisition of Mellanox, April 27, 2020
  • NVIDIA and SoftBank Group, NVIDIA and SoftBank Group Announce Termination of NVIDIA’s Acquisition of Arm Limited, February 7, 2022
  • Hugging Face (Adina Yakefu, Irene Solaiman), One Year Since the “DeepSeek Moment”, January 20, 2026
  • Hugging Face, Optimum-TPU package page (archived), accessed September 5, 2026
  • Clément Delangue, Julien Chaumond, posts on X, September 3, 2026

Reporting and analysis

  • Truth on the Market (Onyeka Aralu, Dirk Auer), Acquihires and Antitrust: When Buying the Team Isn’t Buying the Company, April 9, 2026
  • CNBC, Nvidia agrees to buy Hugging Face for almost $13 billion, AI expansion, September 3, 2026
  • CNBC, Why Nvidia’s Hugging Face deal is about much more than chips, September 4, 2026
  • CNBC, China: Nvidia violated anti-monopoly law, will continue investigation, September 15, 2025
  • Fortune, Hugging Face goes from a ‘scrappy’ startup named after an emoji to $13 billion Nvidia acquisition, September 3, 2026
  • The Implicator, Nvidia Confirms $12.9B Hugging Face Deal, Sets 2027 Close, September 4, 2026

Related articles on this site

  • NVIDIA’s $6 Billion Poolside License Rests on One Investor Letter, August 24, 2026
  • NVIDIA Doubled Revenue and Borrowed $25 Billion, August 28, 2026
  • Four Runs, Four Times It Noticed, Zero Times It Stopped: Anthropic, OpenAI, and a Deadline That Fell on 1 August, August 1, 2026

Dollar figures not attributed to an official source above are media estimates. This piece is a compilation of NVIDIA’s and Hugging Face’s public documents, both companies’ own posts, and news reporting; it does not represent an official statement from NVIDIA, Hugging Face, or any regulator.